Understanding trailing @ processing in SAS input

One of the first things I noticed when I started poking around in SAS code is that the input statement is very powerful, flexible and hence sometimes hard to understand. It can read pretty much anything in to a dataset as long as you tell it what to do.

The use of trailing @s to take control of how SAS advances the input pointer is a powerful technique to read from input files where the data is laid out in non-standard formats. In this blog post, we will try to understand how trailing @ processing works with the help of some infile statement options and the putlog statement to write to the SAS log.

Let’s take this example from the excellent paper The Input Statement: Where It’s @ ¹ - given an put file where the first variable has to be read beginning from a particular column in the input line based on the value of the second variable.

Age  Type
23   1
  44 2

A first pass at trying to do this will result in code like this:

data ages;
  input @6 type $1.;

  if type='1' then
    input @1 age 2.;
  else if type='2' then
    input @3 age 2.;
  drop type;
  datalines;
23   1
  44 2
;
run;

proc print data=ages;
  title "Age read without trailing @";
run;

The result we get is not what we expect.

| Obs | Age |
|-----|-----|
| 1   | .   |

We can understand how SAS read this input using the line= and column= options of the infile statement. While the code in the above listing does not explicitly use infile to point at the datalines, this can be done when we want to use infile’s options with datalines.

This is a good time to remind us what the line= and column= options of infile does. From the documentation:

LINE=variable

specifies a variable that SAS sets to the line location of the input pointer in the input buffer. As with automatic variables, the LINE= variable is not written to the data set.

COLUMN=variable

names a variable that SAS uses to assign the current column location of the input pointer. As with automatic variables, the COLUMN= variable is not written to the data set.

We can modify our code to set these options and then print the value of all variables including these options using the _all_ variable that prints the Program Data Vector (PDV).

data ages;
  infile datalines line=line column=col;

  input @6 type $1.;
  putlog "After reading type, before reading age: " _all_ ;

  if type='1' then
    input @1 age 2.;
  else if type='2' then
    input @3 age 2.;
  putlog "After reading age: " _all_ ;
  putlog "";

  drop type;
  datalines;
23   1
  44 2
;
run;

When the code runs, the following can be seen in the SAS logs:

 After reading type, before reading age: line=1 col=7 type=1 age=. _ERROR_=0 _N_=1
 After reading age: line=2 col=3 type=1 age=. _ERROR_=0 _N_=1
 "

After type is read, we can see that the pointer is at line 1 and column 7, which makes sense considering that the code instructs SAS to go to column 6 and read 1 character. But then we notice that when age is read, the pointer has moved to 2 and column 3 as instructed in the first branch of the if else condition. Since there is nothing at column 1-2 of line 2 in the input, a missing value is stored in age.

What we want is some way to tell input to ot move the pointer after the first input statement. This is where the single trailing @ comes in. It intructs input to stay on the same line for the next input statement in the data step. The above listing modified to use trailing @ is as follows:

data trailing;
  infile datalines line=line column=col;

  input @6 type $1. @;
  putlog "After reading type, before reading age: " _all_ ;

  if type='1' then
    input @1 age 2.;
  else if type='2' then
    input @3 age 2.;
  putlog "After reading age: " _all_ ;
  putlog "";

  drop type;
  datalines;
23   1
  44 2
;
run;

The following log lines are written:

 After reading type, before reading age: line=1 col=7 type=1 age=. _ERROR_=0 _N_=1
 After reading age: line=1 col=3 type=1 age=23 _ERROR_=0 _N_=1
 "
 After reading type, before reading age: line=1 col=7 type=2 age=. _ERROR_=0 _N_=2
 After reading age: line=1 col=5 type=2 age=44 _ERROR_=0 _N_=2
 "

The log shows that input stayed on line 1 even after reading the type variable.

The new dataset has the expected values:

| Obs | Age |
|-----|-----|
| 1   | 23  |
| 2   | 44  |

Here is another example that you often run in to on the internet when discussing trailing @:

data  colors;
  infile datalines line=linenum column=col;
  input @1 Var1 $ @8 Var2 $ @;
  putlog "After reading Var1 and Var2: " _all_;

  input @1 Var3 $ @8 Var4 $ ;
  putlog "After reading Var3 and Var4: " _all_;
  putlog "";

  datalines;
RED    ORANGE  YELLOW  GREEN
BLUE   INDIGO  PURPLE  VIOLET
CYAN   WHITE   FUCSIA  BLACK
GRAY   BROWN   PINK    MAGENTA
;
run;

This results in the following dataset:

|Obs  | Var1 | Var2   | Var3 | Var4   |
|-----|------|--------|------|--------|
| 1   | RED  | ORANGE | RED  | ORANGE |
| 2   | BLUE | INDIGO | BLUE | INDIGO |
| 3   | CYAN | WHITE  | CYAN | WHITE  |
| 4   | GRAY | BROWN  | GRAY | BROWN  |

Because of the trailing @ in the first input statement, the values of Var1 and Var3 for all the observations are the same, as both are read from column 1. Similarly Var2 and Var4 are the same as they are read from column 8.